Capture Solution Speed/Quality Considerations

[jfsecure]

I've been researching and testing our current solution using so many variables my head is spinning.

Here is our current situation.

We have an application that stores our interlaced Panasonic Pro DV footage in file chunks on our main central capture station. The core application only reassembles these pieces in order when it is being played out in analog via the S-video output devices. The video is very fast motion.

We have a need to capture this output and encode it for distribution via the Internet to our other locations. The video is usually 90-120 minutes long and we would like it completely encoded and ready for upload within 6-8 hours of receiving it.

I have a few specific points I'd like to hear some suggestions on.

Since it has to be "played out" using svideo I can't see anyway to avoid capturing and encoding on a seperate capture station in one step. This eliminates the ability to do a multipass encode but I think we can live the file sizes of a single pass encode.

Based on this I have been stuck on which way to go,

1.) Using either a regular ATI Radeon AIW 9XXX series capture card and encoding directly to either xvid or h264/avc

2.) Trying the new ATI X1800 capture card with hardware based GPU support for both encoding and decoding to MP4. I've heard mixed reviews on this and I would have to build all new machines to get the PCI Express support so it would be an expensive test.

3) Trying somethink like a Plextor PX-m402u and living with a minor drop in quality with this purely external device. (Hopefully only minor quality difference)

We are currently leaning towards xvid for for the codec or h.264/avc using the x264 codecs. We will likely use a 640x480 for true (4:3) resolution. There seems to be pretty neglible differences in our situation unless I'm missing something.

I'm really looking forward to some dialogue on this.

[pelmen]

I guess it depends what your end result needs are. Is the final file being played going to be streamed over the net (in which case you have to consider the lowest connection speed and therefore the maximum bitrate for the encoding). If the file is being copied over a network and then played back locally you need to consider that codecs the machines will support and if you should be supplying codecs for playback etc.

An external encoding box solution would probably be easiest but maybe not the best quality. Running the s-video into a DV camcorder and then copying the DV file to your machine is another option. What about a DVD-Recorder unit and capture straight to a DVD? That will give you an instant source backup to archive plus a format you can use to convert to AVI/MP4 with relatively easily.

I'm not sure if going for a hardware based card will help. I use a PVR card which has MPEG2 encoding in hardware but I never use it. Instead I record using VirtualDub and encode straight to divx and the quality is way better plus I can further process the file down to a final output file with much better results than if my source was MPEG2. On a P4-2.6 (no HT and only 400MHz FSB) I can record my divx at balanced setting usually at around 5Mbps without any dropped frames and very little noticable compression artifacts. From there I multipass (sometimes using a deinterlace filter depending on the analogue source I record from) to my final avi. My machine is first generation P4 (socket 423 motherboard with a socket 478 chip/converter), so no PCI-E, hyperthreading or multicore etc. If you put together a machine for encoding you could easily get multipass encoding done as well as realtime capture well within your timeframe. If you put together a machine though I would advise you to make use of a fast drive purely for capturing, that way you avoid capturing to a system or programs drive where the OS may be trying to read data during the capture and you avoid dropped frames from both that and fragmentation issues. Plus a dedicated drive just for capture will get a hell of a lot of work so once your capturing is done all you need to do is reformat and its clean again..no hidden fragments floating around that might slow down writes and cause dropped frames.

An AIW card should be an affordable solution and good enough quality. Analogue video capture is not going to be crystal clear compared to a all digital video pathway so in your case I don't think there is a need for expensive capture cards.

As for filesize I'm pretty sure there is virtually no difference between single and multipass encoding for filesize. It is the bitrates that are the major factors for filesize and encoding passes greatly help the visual quality. Being able to capture at a high bitrate I feel will be the important factor for you (especially with fast motion in the video). Ideally being able to capture to a lossless format would be best to avoid compression artifacts in the capture (there may already be compression artifacts on your capture server so its best to avoid compounding these). Capturing at full resolution at 5Mbps or above should give you a 95+% perfect copy, that will give you the chance to deinterlace/rescale/colour balance with something like VirtualDub as you re-encode to your final file. Getting it into your computer cleanly at a high bitrate will depend on having nothing else running in the background, a fast CPU and a fast hard drive to write to (dedicated hard drive preferrably, though dedicated partition can help but not ideal). Audio captures best at CD quality uncompressed and is easy to deal with in most packages.

Good luck and have fun experimenting!

[signal]

We have an application that stores our interlaced Panasonic Pro DV footage in file chunks on our main central capture station. The core application only reassembles these pieces in order when it is being played out in analog via the S-video output devices.


Sorry, I'm missing something.

I'm a bit confused on why the chunks couldn't be be converted ahead of time on the same station avoiding another analog capture.

Does the core application assemble them in a different order each time?
Is it a manual process?

I was thinking if the material is put together the same way each day, then maybe a simple AVISynth script could be used to assemble them so they look like one avi and dump that through VirtualDub to do the conversion to another codec.

Could even be batched so it's automated or a one click process.

[signal]

Guess jfsecure wasn't that interested

[The_Fugitive]

@jfsecure:
The speedy solution would be to use the Plextor device you mentioned, since it outputs the capture in realtime to the MPEG format, (also DivX) only for you to judge if quality is up to your standard, internal capture cards are for sure to give qualty loss, the other option is to take a look on the Canopus site, it offers a wide range of options, from amateur to Pro.